Design and Application of a Dual-stream Convolutional Neural Network Integrating Expert Experience and Transfer Learning

This paper constructs an end-to-end trained dual-stream convolutional neural network model. The model consists of two channels. The first channel adopted a deep convolutional neural network which achieved impressive performance on the Imagenet dataset. On the basis of the network, the transfer learning method is exploited to resolve the problems caused by insufficient datasets, which can ensure the performance of the model. The features extracted through two channels are fused at the decision-making level and then input into the fully connected network for classification.